Save successfully generated code as skills | Open Interpreter | Page 1

paper sun Sep 11, 2023, 5:08 PM

#

Sometimes it takes several iterations to get the correct script/code. But once you do, it would be lovely to be able to invoke it again. Example of this is show me upcoming birthday's this week.

cold moon Sep 11, 2023, 5:10 PM

#

Sound interesting.

chrome ermine Sep 11, 2023, 5:11 PM

#

adding my comment here, as it's relevant. this could be stored in a local version of open-procedures #feature-ideas message

obtuse sundial Sep 13, 2023, 9:12 AM

#

paper sun Sometimes it takes several iterations to get the correct script/code. But once y...

it may be like flowerise to save generate code to execute flow, may an thrid party app do this more suitable

sullen tide Sep 13, 2023, 11:26 AM

#

plz also see my comments here #feature-ideas message
and I also write a blog to talk about the details of the implement: https://granite-agustinia-12c.notion.site/Open-creator-in-open-interpreter-Why-How-What-and-When-ff05a2be4f62450b980ac862f2eee288?pvs=4

俊民龚's Notion on Notion

Open creator in open interpreter - Why, How What, and When?

Created by @chuxij & @chatgpt

primal frigate Sep 13, 2023, 3:19 PM

#

Very interesting idea

tender cliff Sep 14, 2023, 1:19 AM

#

this is a great idea

cold moon Sep 14, 2023, 2:06 PM

#

a pull request has been created for load the stored message when the OI start. It is not a full solution, but it would save some of the back and forward conversation by using the message history.

https://github.com/KillianLucas/open-interpreter/pull/359

GitHub

Feature: load message from command line by moming2k · Pull Request ...

Pull Request Overview
This pull request introduces a feature that allows the application to load previously saved messages upon startup. This enhancement allows the reuse of prior knowledge, elimin...

sullen tide Sep 16, 2023, 4:38 AM

#

%save_skill
+
%load_skill

pull request: https://github.com/KillianLucas/open-interpreter/pull/399

GitHub

feat: add %save_skill & %load_skill command by ChuxiJ · Pull Reques...

Feat: add %save_skill and %load_skill

Extract your conversation history into a skill object using the reflection agent

You can save it to a path or default skill library

You can load it fr...

spare echo Sep 16, 2023, 10:27 AM

#

Excellent work, still reading.

#

This looks suspiciously like it would fit in a knowledge graph.

spare echo Sep 16, 2023, 10:54 AM

#

Trying PR 399 @sullen tide
llama-cpp-python errors, again.

gh repo clone KillianLucas/open-interpreter
cd open-interpreter
gh pr checkout 399

virtualized
poetry install
poetry run interpreter --local

Building wheels for collected packages: llama-cpp-python
  Building wheel for llama-cpp-python (pyproject.toml) ... error
  error: subprocess-exited-with-error

  × Building wheel for llama-cpp-python (pyproject.toml) did not run successfully.
  │ exit code: 1
  ╰─> [47 lines of output]
      *** scikit-build-core 0.5.0 using CMake 3.27.4 (wheel)
      *** Configuring CMake...
      2023-09-16 06:48:00,524 - scikit_build_core - WARNING - libdir/ldlibrary: /home/kyle/anaconda3/lib/libpython3.10.a is not a real file!
      2023-09-16 06:48:00,524 - scikit_build_core - WARNING - Can't find a Python library, got libdir=/home/kyle/anaconda3/lib, ldlibrary=libpython3.10.a, multiarch=x86_64-linux-gnu, masd=None

#

pip install llama-cpp-python

Building wheels for collected packages: llama-cpp-python
  Building wheel for llama-cpp-python (pyproject.toml) ... error
  error: subprocess-exited-with-error

  × Building wheel for llama-cpp-python (pyproject.toml) did not run successfully.
  │ exit code: 1
  ╰─> [66 lines of output]
      *** scikit-build-core 0.5.0 using CMake 3.27.4 (wheel)
      *** Configuring CMake...
      2023-09-16 06:55:49,846 - scikit_build_core - WARNING - libdir/ldlibrary: /home/kyle/anaconda3/lib/libpython3.10.a is not a real file!
      2023-09-16 06:55:49,847 - scikit_build_core - WARNING - Can't find a Python library, got libdir=/home/kyle/anaconda3/lib, ldlibrary=libpython3.10.a, multiarch=x86_64-linux-gnu, masd=None

sullen tide Sep 16, 2023, 10:57 AM

#

spare echo Trying PR 399 <@1037535679065903104> llama-cpp-python errors, again. ``` gh repo...

just ignore. not use llama-cpp. only test for gpt-4

spare echo Sep 16, 2023, 10:57 AM

#

Is this my environment screwing things up?

#

ok, then I need to hack in the litellm keys right?

#

I don't have tokens

#

pr 266 claude

#

er 288

#

Seeing if magic keys work

sullen tide Sep 16, 2023, 11:02 AM

#

open router may be an option or other proxy server of openai

spare echo Sep 16, 2023, 11:22 AM

#

Skeleton key: activate! Got it running

#

  The function extract_skill_from_conversation has been defined. This function takes a conversation history as input and returns the first skill object it finds in the
  conversation. A skill object is a message that includes all the properties mentioned in the 'extract_formmated_skill' function.

  Next, we need a conversation history to test this function. Could you please provide a conversation history?

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/kyle/src/github/skillz/open-interpreter/interpreter/interpreter.py", line 141, in cli
    cli(self)
  File "/home/kyle/src/github/skillz/open-interpreter/interpreter/cli.py", line 207, in cli
    interpreter.chat()
  File "/home/kyle/src/github/skillz/open-interpreter/interpreter/interpreter.py", line 478, in chat
    self.handle_command(user_input)
  File "/home/kyle/src/github/skillz/open-interpreter/interpreter/interpreter.py", line 380, in handle_command
    action(arguments)  # Execute the function
  File "/home/kyle/src/github/skillz/open-interpreter/interpreter/interpreter.py", line 319, in handle_save_skill
    json_path = save_skill_to_library(skill_json, json_path)
  File "/home/kyle/src/github/skillz/open-interpreter/interpreter/skill_extractor.py", line 112, in save_skill_to_library
    save_path = default_skill_lib_path + skill_json['skill_name'] + ".json"
KeyError: 'skill_name'

#

You were doin' great Kid, right up until you blew up. -- Iron Eagle

#

Going to kill some pixels. Be back after lil gaming.

sullen tide Sep 16, 2023, 11:39 AM

#

spare echo You were doin' great Kid, right up until you blew up. -- Iron Eagle

Normally we consider above “messages “ as conversation history🤯. I need to do more tests to its robustness.

#

Please use gpt-4 to test it

#

You can run a common request first and then save it as a skill

spare echo Sep 16, 2023, 11:45 AM

#

That was gpt-4

#

write a program to list all the python functions and their docstrings in a directory

sullen tide Sep 16, 2023, 11:51 AM

#

spare echo ``` write a program to list all the python functions and their docstrings in a d...

thank you for your feedback! let me figure it out

sullen tide Sep 16, 2023, 3:00 PM

#

spare echo ``` The function extract_skill_from_conversation has been defined. This functi...

problem fixed. See the latest commits. this is because there are conflicts between skill_extractor system prompt and open-interpreter system prompt.

I am going to write an independent package called open-creator to construct skill objects.

spare echo Sep 16, 2023, 4:37 PM

#

Is there something you'd like tested? @sullen tide

sullen tide Sep 17, 2023, 12:07 AM

#

spare echo Is there something you'd like tested? <@1037535679065903104>

Yep, test how to create an agent that it can create itself

spare echo Sep 17, 2023, 12:32 AM

#

This right:

gh repo clone KillianLucas/open-interpreter
cd open-interpreter
gh pr checkout 399

virtualized

#

📎 message.txt

#

By George, I think you've got it!

#

head /home/kyle/.cache/open_interpreter/skill_library/list_python_functions_and_docstrings.json
{
  "skill_name": "list_python_functions_and_docstrings",
  "skill_description": "This skill lists all the Python functions and their docstrings in a specified directory. It uses the 'os' module to list all Python files in the directory, and the 'ast' module to parse each file and find all function definitions. For each function, it extracts its name and docstring.",
  "skill_dependencies": [
    "os",
    "ast"
  ],
  "skill_parameters": [
    {
      "param_name": "directory",

#

Looks like it loads too

#

Sooo, once it's loaded, what do I do? How do I invoke

sullen tide Sep 17, 2023, 12:51 AM

#

spare echo Sooo, once it's loaded, what do I do? How do I invoke

you can reuse this skill by giving parameters

spare echo Sep 17, 2023, 12:51 AM

#

What do I type?

#

Is this supposed to be runnable from the loop?

#

I just call it?

sullen tide Sep 17, 2023, 12:53 AM

#

you try to say “please list the python function from xxx” xxx is your code path

spare echo Sep 17, 2023, 12:54 AM

#

Not sure what this did, ok

list_python_functions_and_docstrings('interpreter')


  Output truncated. Showing the last 2000 characters.

  ase names and combines the sizes and RAM requirements.\n\n:param models: List of model details.\n:return: A list of combined model details.'), ('actually_combine_files',
  'Combines files together and deletes the original split files.\n\n:param base_name: The base name for the combined file.\n:param files: List of files to be combined.'),
  ('format_quality_choice', 'Formats the model choice fo

#

Locked it up, yes!

#

I break things real good, heh, restarting

sullen tide Sep 17, 2023, 12:55 AM

#

spare echo Is this supposed to be runnable from the loop?

yes in the loop and it runs

spare echo Sep 17, 2023, 12:56 AM

#

oic

#

I noticed when generating it flickered at me

#

please list all the python functions in interpreter

  Sure, I will list all the Python functions in the interpreter. To do this, I will use the built-in Python function dir() which returns a list of names in the current
  local scope. I will filter this list to only include callable objects, which are the functions.

  Let's execute this plan.



  import types

  functions = [f for f in dir() if callable(globals()[f]) and isinstance(globals()[f], types.FunctionType)]
  functions


  Would you like to run this code? (y/n)

#

You should spin something to tell people it's not stuck

sullen tide Sep 17, 2023, 12:57 AM

#

Yes you’re right

spare echo Sep 17, 2023, 12:58 AM

#

please list all the python functions in the interpreter directory

  To list all the Python functions in the interpreter directory, we need to:

   1 List all the Python files in the directory.
   2 For each Python file, import it as a module.
   3 List all the functions in each module.

  Let's start with the first step: listing all the Python files in the directory.



  import os

  python_files = [f for f in os.listdir() if f.endswith('.py')]
  python_files


  Would you like to run this code? (y/n)

#

This doesn't look like it's using the function

sullen tide Sep 17, 2023, 12:59 AM

#

I didn’t add the functionality of semantic search

#

You need load the skill first by hand

spare echo Sep 17, 2023, 1:00 AM

#

I loaded the skill

#

let me make sure, there's no way to list the ones loaded

sullen tide Sep 17, 2023, 1:01 AM

#

Seems like we need more development

spare echo Sep 17, 2023, 1:01 AM

#

Ya, I had it loaded, it loops trying to import. It finally gets to where there's no output and loops

#

It tries to solve it various ways

#

Lemee videe

#

oic, my autoenv is killing it with terminal messages. sec

#

I removed my .env file that was causing errors

sullen tide Sep 17, 2023, 1:07 AM

#

spare echo oic, my autoenv is killing it with terminal messages. sec

My fault. There’s a bug maybe… I didn’t pass the skill example to the loop

spare echo Sep 17, 2023, 1:08 AM

#

It workled

#

wonder how long a vid you can post guess I'll see

sullen tide Sep 17, 2023, 1:10 AM

#

spare echo Sep 17, 2023, 1:10 AM

#

Too big.: output:

  Here are the functions in the interpreter module along with their docstrings:

   • __class__: No docstring provided.
   • __delattr__: Implement delattr(self, name).
   • __dir__: Default dir() implementation.
   • __eq__: Return self==value.
   • __format__: Default object formatter.
   • __ge__: Return self>=value.
   • __getattribute__: Return getattr(self, name).
   • __gt__: Return self>value.
   • __hash__: Return hash(self).
   • __init__: No docstring provided.
   • __init_subclass__: This method is called when a class is subclassed. The default implementation does nothing. It may be overridden to extend subclasses.
   • __le__: Return self<=value.
   • __lt__: Return self<value.
   • __ne__: Return self!=value.
   • __new__: Create and return a new object.  See help(type) for accurate signature.
   • handle_save_message: No docstring provided.
   • handle_save_skill: No docstring provided.
   • handle_undo: No docstring provided.
   • load: No docstring provided.
   • reset: Resets the interpreter.
   • respond: No docstring provided.
   • verify_api_key: Makes sure we have an AZURE_API_KEY or OPENAI_API_KEY.

sullen tide Sep 17, 2023, 1:11 AM

#

I don't know. the open-interpreter didn't use the skill I provided to solve the problem. This a big issue.

spare echo Sep 17, 2023, 1:11 AM

#

oh ya lol

#

it should have right?

#

let me ask it

#

Nein, mein Herr, it is kerplunked:

show me all the functions with their docstring in the interpreter directory


  import os
  import importlib

  # Get all python files in the directory
  py_files = [f[:-3] for f in os.listdir('.') if f.endswith('.py')]

  # Import all modules
  modules = [importlib.import_module(f) for f in py_files]

  # Get all functions and their docstrings
  func_docstrings = {}
  for module in modules:
      funcs = [func for func in dir(module) if callable(getattr(module, func))]
      for func in funcs:
          func_docstrings[f'{module.__name__}.{func}'] = getattr(module, func).__doc__

  func_docstrings


  Would you like to run this code? (y/n)

#

Who knew it's name was Shcultz?

#

%list_skills

sullen tide Sep 17, 2023, 1:14 AM

#

just %load_skill, it will give you the skill list in skill library

sullen tide Sep 17, 2023, 1:23 AM

#

sullen tide I don't know. the open-interpreter didn't use the skill I provided to solve the ...

I am fixing it. I will create a new commit for this issue.

spare echo Sep 17, 2023, 1:24 AM

#

Logical, I'm taaking a little walk, test if your ready when back

sullen tide Sep 17, 2023, 1:32 AM

#

solved! See the latest commits

#

here is the main change

spare echo Sep 17, 2023, 3:02 AM

#

Voodoo to patch to use litellm servers:

sed -i '/None/{s//"litellm-api-key"/;:p;n;bp}' interpreter/interpreter.py
sed -i '/None/{s//"https:\/\/proxy\.litellm\.ai"/;:p;n;bp}' interpreter/interpreter.py

#

This replaces the first occurence of None each time

#

testing

#

Happy mistake, I told it a dir. I'm a wrecking ball! 😉 I'll try with real fpath.

%save_skill /home/kyle/src/github

▌ I am going to extract a skill json object from the above conversation history.

  Would you like to run this code? (y/n)

  y


Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/kyle/src/github/testskillz/open-interpreter/interpreter/interpreter.py", line 139, in cli
    cli(self)
  File "/home/kyle/src/github/testskillz/open-interpreter/interpreter/cli.py", line 207, in cli
    interpreter.chat()
  File "/home/kyle/src/github/testskillz/open-interpreter/interpreter/interpreter.py", line 494, in chat
    self.handle_command(user_input)
  File "/home/kyle/src/github/testskillz/open-interpreter/interpreter/interpreter.py", line 396, in handle_command
    action(arguments)  # Execute the function
  File "/home/kyle/src/github/testskillz/open-interpreter/interpreter/interpreter.py", line 325, in handle_save_skill
    json_path = save_skill_to_library(skill_json, json_path)
  File "/home/kyle/src/github/testskillz/open-interpreter/interpreter/skill_extractor.py", line 116, in save_skill_to_library
    with open(save_path, 'w') as f:
IsADirectoryError: [Errno 21] Is a directory: '/home/kyle/src/github'

#

You get to pick a file extention for .skil files 😉

#

Well, it didn't try to redo a program and just spat it out.

#

Houston, I think we have liftoff

#

list the functions and their docstrings in interpreter

  Here are the functions and their docstrings found in each Python file in the 'interpreter' directory:

   • 'skill_extractor.py':
      • 'save_skill_to_library': No docstring
      • 'get_skill_from_library': "list all the skills in the skill library\n:return: a list of skills"
      • 'print_format_skill': No docstring
   • 'code_block.py':
      • 'get_code': No docstring

sullen tide Sep 17, 2023, 3:29 AM

#

spare echo You get to pick a file extention for .skil files 😉

Yes. If you provide a dictionary path, not json file path, it is supposed to create a json file in that directory
Let me add this logic

spare echo Sep 17, 2023, 3:31 AM

#

Impressive man, great job

sullen tide Sep 17, 2023, 3:36 AM

#

sullen tide Yes. If you provide a dictionary path, not json file path, it is supposed to cre...

done

full heron Sep 20, 2023, 5:41 AM

#

hi @sullen tide , appreciate your PR! it's so cool to build a reusable skill library of fundamental and frequent skills, and I can image how these skills could be potentially stacked to build large and complex project from your excellent writings. Great work🔥

paper sun Sep 20, 2023, 7:16 PM

#

Love seeing this idea come to fruition @sullen tide 🔥🔥🔥🔥🔥🔥

spare echo Sep 20, 2023, 7:19 PM

#

I can't wait to use this

sullen tide Sep 21, 2023, 12:08 AM

#

Current Progress

We are setting up a repository on Space named skill-library. Inside it, we will store the following:

A Gradio app to display the skill-library readme, filtering options (tag, repo_id, skill_name, version), as well as search functionality.
Specific skill directories will look as follows:

(base) ➜  skill_library pwd
/Users/gongjunmin/.cache/open_creator/skill_library
(base) ➜  skill_library tree extract_pdf_section
extract_pdf_section
├── conversation_history.json
├── embedding_text.txt
├── function_call.json
├── install_dependencies.sh
├── skill.json
├── skill_code.py
└── skill_doc.md
1 directory, 7 files

When a user filters or selects a particular skill-library, a script will automatically generate a readme which includes the skill's name, description, usage instructions, and how to import it through the creator.create interface.

import creator
skill = creator.create(huggingface_repo_id="Timedomain-tech/skill-library", huggingface_skill_path="ChuxiJ/skill-library/get_whether", version="1.0.0")

User Upload Process for their Skill Library

The following script is used for saving:
If no repo exists, it will automatically clone our template Timedomain-tech/open-creator-skill-library and create one.
Finally, the skill's JSON files and related content are uploaded.

import creator
creator.save(skill, huggingface_repo_id="ChuxiJ/open-creator-skill-library")

Mechanism for User Skill Submissions

After upload, a pull request will automatically be created in our skill-library git repository, aiming to edit the skill-library.md file.
We will have an automated bot script that will evaluate the submitted skill and conduct an automatic review.
If it passes, it will be automatically merged into the repository.
For the skill-library on Hugging Face Space, updates will be pulled from the skill-library.md file regularly, for instance, daily.

spare echo Sep 21, 2023, 12:13 AM

#

gimme!

#

😉

sullen tide Sep 21, 2023, 10:00 AM

#

The latest progress

Normally, we want to create a skill object after an open-interpreter session (see the pull request 399) . Thanks to the skill_extractor_agent, we can use our conversation history and convert it into a formatted skill object by using %save_skill command.

But now, we can create a skill object by using only a request.

#

How this happened?

We fork the code of open-interpreter , refactor and re-implement it into a minimal version

Code interpreter works in only 45 lines has been integrated into open-creator

import subprocess
import traceback


def get_persistent_process(start_cmd: str):
    process = subprocess.Popen(
        args=start_cmd.split(),
        stdin=subprocess.PIPE,
        stdout=subprocess.PIPE,
        stderr=subprocess.PIPE,
        text=True,
        bufsize=0
    )
    return process


class BaseInterpreter:
    """A tool for running base code in a terminal."""

    name: str = "base_interpreter"
    description: str = (
        "A base shell command tool. Use this to execute bash commands. "
        "It can also be used to execute any language with interactive mode"
    )

    def __init__(self):
        self.process = None

    def run(self, query: str, is_start: bool = False) -> dict:
        if is_start or self.process is None:
            try:
                self.process = get_persistent_process(query)
                return {"status": "success", "stdout": "", "stderr": ""}
            except Exception:
                traceback_string = traceback.format_exc()
                return {"status": "error", "stdout": "", "stderr": traceback_string}
        
        stdout, stderr = "", ""
        try:
            stdout, stderr = self.process.communicate(input=query)
        except BrokenPipeError:
            stderr = traceback.format_exc()

        return {"status": "success", "stdout": stdout, "stderr": stderr}

(open_creator_dev) ➜  code_interpreter git:(main) ✗ tree .                         
.
├── R.py
├── __init__.py
├── applescript.py
├── base.py
├── html.py
├── javascript.py
├── julia.py
├── python.py
└── shell.py

#

Like chatDev, GPT-Enginer, GPT-team and etc, we can use autonomous agent. There will be agents not only help you write code, converting codes into skills, but also doing the tests, and the refactor stuff...

Do you have any good ideas for this?

spare echo Sep 21, 2023, 10:29 AM

#

I'm going to put that in my bootcupboard named skill_extractor_agent.py to play with if anyone wants to get in future.
Yoink!

#

uploaded

#

Thank you, thank you very much. -- Elvis

#

You mean how to use this in refactoring and testing this project?

#

@sullen tide

sullen tide Sep 21, 2023, 10:36 AM

#

yes, how to design refactoring and testing agent so we can have more complicate skill examples
for example using langchain and its api documentation to build some new type of stuffs

spare echo Sep 21, 2023, 10:40 AM

#

Cool, I will run it and see if I can blow it up 😉

#

and see if I can think of a proof of concept

#

Thank you

#

I'm going to steal get_persistent_process() for other things too. so many toys

#

I was going to have to write something like this for other things I was doing. Very handy and well written.

sullen tide Sep 21, 2023, 10:46 AM

#

Wonderful

paper sun Sep 21, 2023, 9:29 PM

#

One request I have created a bunch of skills in applescript already. two quick questions 1. Can they be loaded into this? 2. Is it possible to use LLM to figure out which script to invoke based on the question? cc @cinder axle

sullen tide Sep 21, 2023, 11:51 PM

#

paper sun One request I have created a bunch of skills in applescript already. two quick...

this is what open creator does,

we still need a discussion after pre-release. how to integrate into open-interpreter

spare echo Sep 21, 2023, 11:55 PM

#

Like this? https://discord.com/channels/1146610656779440188/1151597896072958094

#

#

The rings are neo4j Graph DBs The pyramid is a vector store matcher

#

Text is asosciated with the tool / node and VDB is used to match

sullen tide Sep 22, 2023, 12:05 AM

#

we simply use langchain tool selection

https://python.langchain.com/docs/modules/agents/how_to/custom_agent_with_tool_retrieval

Custom agent with tool retrieval | 🦜️🔗 Langchain

This notebook builds off of this notebook and assumes familiarity with how agents work.

spare echo Sep 22, 2023, 12:09 AM

#

I've looked at that.

#

If it works great. No idea how well

#

I'll take another look and see how hard a proof of concept would be

spare echo Sep 22, 2023, 12:27 AM

#

Our Agent springs into action!

spare echo Sep 22, 2023, 1:23 AM

#

https://discord.com/channels/1146610656779440188/1154577941204828160

sullen tide Sep 22, 2023, 1:33 AM

#

📣 Code Refactoring Agent Ideas 🛠️

Transformations:
- Add docstrings automatically.
- Auto-rename functions and variables based on naming conventions.
- Simplify complex logic.
- Introduce callbacks or error handling mechanisms.
Splitting:
- Break down large code blocks or functions into smaller, more specific parts for better readability and maintainability.
Assembly:
- Combine smaller code snippets or functions into a larger unified piece, especially useful when there are repetitive code segments.

💡 Using a simple language to orchestrate these refactoring skills:

Combine two skills with user input: skillC = skillA + skillB => user_request
Split one skill into multiple: skill_list = skillA => user_request
Assemble multiple parts into one: skill = skill_list => user_request
Chain skills: skill = skillA - skillB - skillC (can be no user request input)

What do you think would be the biggest challenges in implementing such a refactoring agent, and how would you address them? Let's brainstorm together! 🧠💬

https://discord.com/channels/1146610656779440188/1154663621352230922

radiant anchor Sep 22, 2023, 3:52 AM

#

Just found this thread. Super interesting and useful. Let me know if I can help!

spare echo Sep 22, 2023, 4:13 AM

#

@radiant anchor There's a lot to go through. Any help is welcome. I have such a list, like Santa

radiant anchor Sep 22, 2023, 7:42 AM

#

So I'm clear, whats actually functional here?

#

And, whats the compressed version of the wishlist?

burnt gorge Sep 22, 2023, 10:03 AM

#

I was wondering how you'd deal with all the nuances of very similar requests?
For example
"[Very complex instruction]"
is very similar to
"[Very complex instruction] then filter it by x"
And would probably get detected as the skill.
I think you'd have to save and embed each step aswell as the code

#

I've not managed to look at the code properly yet so I'm not sure how much further you've got, I was just thinking about it

#

IMO 'Skills' wouldn't be the same as agent 'tools' or 'actions'.
Any set of steps that can be hardcoded shouldn't be a skill, and anything that is infeasible to hardcode steps should be a skill

#

Like playing basketball is a skill, but telling the time is not

#

I think all of this put together would make a really useful agent

#

You would need to stop it looking for skills while it's already doing one

#

Unless you have recursive skills

#

Or would each step itself be a skill?

sullen tide Sep 22, 2023, 11:55 PM

#

@burnt gorge thanks your invaluable feedback! It’s really a great thought. We need to think how to narrow the scope of the skills and avoid to generalize its definition too much.
Currently we prefer to consider some shadow help utils code functions or procedure tips as skills

We currently focus on using automatic ways to create skill objects and make it persistent so that it can be easily reused

Your concern is really important

The concept or schema of Skills is not totally the same with tools or actions

Skills here we talk can be created by agents and optimized by human and test agents but tools are normally man crafted

Also actions sound like a function call it means a function selection in the agent’s action space but here our actions to agents are "search skills”,“create a skill”, “test the skill”, and “refactoring skills”

#

We planed to have a pre-release this weekend. Maybe 2-3 days delay at most

I have finished the overall project but still have more tests and documentation stuff to do

Plz stay tuned 😉

spare echo Sep 23, 2023, 11:42 AM

#

#feature-ideas message

#

Use local model to look for saved skill and use that if found instead of using the large model to find the skill?

burnt gorge Sep 24, 2023, 12:01 AM

#

Sounds interesting look forward to seeing how it goes

loud garnet Sep 24, 2023, 3:14 AM

#

follow big time🚀🚀🚀

burnt gorge Sep 24, 2023, 4:26 AM

#

You could make them universal by telling it to generate cross platform code, then people could share them, but how would you deal with where the code contained personal info like a specific file path, and how would you know which parts of the code should be changed when the skill is rerun

#

Maybe tell it to generate code with placeholders, then when you come to run it replace the placeholders?

sullen tide Sep 24, 2023, 5:56 AM

#

Normal skills are functions or classes that can be externally imported

sullen tide Sep 28, 2023, 5:23 PM

#

pre-prelease see: https://github.com/timedomain-tech/open-creator
we still have a lot of tests to do

GitHub

GitHub - timedomain-tech/open-creator

Contribute to timedomain-tech/open-creator development by creating an account on GitHub.

chrome ermine Oct 2, 2023, 2:39 AM

#

sullen tide pre-prelease see: https://github.com/timedomain-tech/open-creator we still have ...

Description: "Build your costomized skill library An open-source LLM tool helps create your tools"

spelling/grammar edits: "An open-source LLM tool for extracting repeatable tasks from your conversations, and saving them into a customized skill library for retrieval."

sullen tide Oct 2, 2023, 2:51 AM

#

chrome ermine Description: "Build your costomized skill library An open-source LLM tool helps ...

Thanks, let me update this

chrome ermine Oct 2, 2023, 2:51 AM

#

thank you for your work! excited to try when i have a minute. a little copy edit is the least i could do 😄

sullen tide Oct 2, 2023, 2:53 AM

#

Done!

chrome ermine Oct 2, 2023, 2:53 AM

#

also, " ~~Sreaming~~Streaming: Stream your function calls"

#

At top of main description

Build your ~~costomized~~customized skill library

sullen tide Oct 2, 2023, 2:55 AM

#

Yes yes yes

spare echo Oct 2, 2023, 10:00 AM

#

@sullen tide I have some time before I go to neo4j-land to do tests. You said refactor and test agents last.

sullen tide Oct 2, 2023, 12:14 PM

#

spare echo <@1037535679065903104> I have some time before I go to neo4j-land to do tests. Y...

remember to pull the latest code, there are some bugs fixed

spare echo Oct 2, 2023, 12:23 PM

#

Ok, somebody had an issue with its load function yesterday

#

I should use pip or repo?

#

Trying to install repo:

Installing dependencies from lock file
Warning: poetry.lock is not consistent with pyproject.toml. You may be getting improper dependencies. Run `poetry lock [--no-update]` to fix it.

Because open-creator depends on litellm (^0.1.590) which doesn't match any versions, version solving failed.

#

Fix #1:

pip install litellm

#

Because open-creator depends on inquirer (^3.1.3) which doesn't match any versions, version solving failed.

and so on, manually installing deps

#

prompt_toolkit

#

inquirer

#

last dep

#

  • Installing annotated-types (0.5.0)
  • Downgrading charset-normalizer (3.3.0 -> 3.2.0)
  • Installing mypy-extensions (1.0.0)
  • Downgrading packaging (23.2 -> 23.1)
  • Installing pydantic-core (2.6.3)
  • Installing greenlet (2.0.2)
  • Installing marshmallow (3.20.1)
  • Installing mdurl (0.1.2)
  • Installing numpy (1.25.2)
  • Installing pydantic (2.3.0)
  • Installing typing-inspect (0.9.0)
  • Installing dataclasses-json (0.6.0)
  • Installing exceptiongroup (1.1.3)
  • Downgrading fsspec (2023.9.2 -> 2023.9.1)
  • Installing iniconfig (2.0.0)
  • Installing langsmith (0.0.40)
  • Installing markdown-it-py (3.0.0)
  • Installing numexpr (2.8.6)
  • Installing pluggy (1.3.0)
  • Installing pygments (2.16.1)
  • Installing sqlalchemy (2.0.21)
  • Installing tenacity (8.2.3)
  • Updating huggingface-hub (0.16.4 -> 0.17.2)
  • Installing langchain (0.0.297)
  • Installing loguru (0.7.2)
  • Downgrading openai (0.28.1 -> 0.28.0)
  • Installing pytest (7.4.2)
  • Installing rich (13.5.3)
  • Downgrading tiktoken (0.5.1 -> 0.4.0)

#

That markdown lib is nifty. I want to rewrite my debugging to use it. Make it all purty and in a fancy getup
from rich.markdown import Markdown

#

I can't auth with openai. Nothing that uses that will work for me.

openai.error.AuthenticationError: Incorrect API key provided: openinte*******-key. You can find your API key at https://platform.openai.com/account/api-keys.

#

python3 ./test_tester.py

spare echo Oct 2, 2023, 1:32 PM

#

@sullen tide I hope you don't mind. I was reading your paper. Awesome, by the way, extremely impressive. I think the wording in the abstract could be clearer.

#

I'll finish later. So is there anything which doesn't require a valid openai key?

spare echo Oct 2, 2023, 2:29 PM

#

There are a few other word choices if you like, just smoother ways of saying things sometimes.

#

leading to low robustness = leading to fragility for example

sullen tide Oct 3, 2023, 2:16 AM

#

Thank you for your valuable suggestion. I am diligently compiling all of your advice, encompassing not only code-related issues but also including matters pertaining to spelling, grammar, and word rephrasing.

spare echo Oct 3, 2023, 4:04 AM

#

I hope you don't mind

#

Love the paper

#

So, I don't have an openai key . I tried running some code and got that error. Is this a requirement or is there some way to reconfigure?

sullen tide Oct 3, 2023, 4:33 AM

#

spare echo So, I don't have an openai key . I tried running some code and got that error. I...

Let me give you one

spare echo Oct 3, 2023, 5:27 AM

#

I just did a complete system reinstall and dropped to ubuntu 22.04 branch. 23.04 was causing me CUDA issues. Going to try nix for environmental compartamentalization

untold ember Oct 3, 2023, 2:23 PM

#

Hi everyone, I've been silently following open-interpreter for a while and I wonder if it is overlooked that gpt-3.5-turbo-0613 and gpt-4-0613 are fine-tuned for function calling, cause I can imagine these models would be more suitable for the project.

This is straight from the openai website and think it is still valid: "gpt-3.5-turbo-0613 and gpt-4-0613) have been fine-tuned to both detect when a function should to be called (depending on the input) and to respond with JSON that adheres to the function signature."

Nevermind; #general message

#Save successfully generated code as skills

The latest progress